Hi Everyone,
We have been purchasing all of our hosts with dual cpu's (quad core) (around 35 hosts now). We average less then 15% cpu utilization on most all of them. I have been contemplating only purchasing single CPU servers in the future, especially now with six and eight core cpu's. My question is though, just because the actual cpu utilization is low, how would this effect the performance of vm's with multiple vcpu's on a host with a single physical cpu? We have many due to the applications they run (tomcat, apache, java). Just curious about latency or anything with scheduler and the fact that the host has less cores/cpu's then all combined vm's with multiple vcpu. We average about 30-40 vm's per server. 75% are 2 vcpu's.
You can use 4 to 6 single vCPU guest per Core
So on a Dual/Quad core (8 vCPU) host,
you can average about 32 to 48 Single core VM's..
you can average about 16 to 24 Dual core VM's
The CPU power is limited by the number of physical cores so if you have VMs with a huge number of vCPUs assigned, the scheduler can only
assign computing cycles if it finds an adequate number of idle physical cores.
So there will be no real negative impact or latency of the vm's waiting on the scheduler to handle multiple instructions for multi vcpu vm's with four less cores on the host? I was always under the impression that just because the cpu resources themselves are not showing high utilization does not mean that the additional cores does not reduce latency for vm's with multiple virtual cpu's. This is what I am trying to understand. Whether there would be a drastic or at least noticeable difference if I reduced to four cores from eight.
Trust the scheduler. The CPU scheduler is aware of all these components and together with the global scheduler (DRS) it can see to it that the virtual machine will receive its resource entitlement.
You can use certain tools to verify performance of CPU in a day or a week and later reduce the no of cores and verify the difference.
